Claude Models on Google Cloud: Anthropic’s managed API arrives

Companies can reach the Claude API from their Google Cloud project, with Google billing and access controls.

On 22 October 2026 Anthropic launched Claude Models on Google Cloud (Anthropic managed). Claude was already available on Google Cloud: the news is a version run by Anthropic itself. Companies can now reach the Claude API from their own Google Cloud project, with access handled by Google Cloud IAM and billing through Google Cloud Marketplace.

Key facts

Launched on 22 October 2026. Claude was already on Google Cloud: this adds a version managed by Anthropic.
It is the Claude API operated by Anthropic, reached through a Google Cloud project.
Access is controlled with Google Cloud IAM.
Billing goes through Google Cloud Marketplace.
Fast mode for Claude Opus 5.5 is available on this service.

What happened

Claude models were already offered on Google Cloud through Google’s own managed service. What is new is a second option. With Claude Models on Google Cloud (Anthropic managed), the service is run by Anthropic, but companies reach it from inside their own Google Cloud project. Who can use it is decided with Google Cloud IAM, the same permission system teams already use for their other cloud resources, and the cost appears on the Google Cloud bill through the Marketplace.

At launch, Anthropic also made fast mode for Claude Opus 5.5 available on this service. Fast mode had left its research preview for Opus 5.5 a week earlier, on 15 October, according to the same release notes.

The launch sits alongside the other places where companies can already run Claude, such as the Claude API itself, Amazon Bedrock and Microsoft Foundry, which Anthropic lists in its release notes for its models.

Why it matters

For many companies the hardest part of adopting a new AI tool is not the technology but the paperwork: a new supplier, a new contract, a new invoice, a new security review. If a team already runs on Google Cloud, using Claude through the same project, the same permissions and the same bill removes much of that friction.

For everyday users inside those companies, this often means Claude shows up sooner in the tools they already use, approved by IT. The prompts should not change: a good prompt written for Claude should work the same way wherever the model runs. That is why it is worth building a small personal library of prompts that you trust, and reusing them as tools arrive in your workplace.

How to try it

A classic first job for AI at work is turning a table of numbers into a clear monthly report. Our free prompt Monthly report from a table is built for that. Fill the brackets with the period, a short description of the business, and paste the table (for example, this month and last month for sales, costs and new customers).

A filled request could start like this: “Period: September vs August. Business: small online shop. Table: sales 12,400 vs 11,260, costs 7,900 vs 7,190, new customers 61 vs 70.” Use your real figures, not these placeholders.

Paste it into Claude or any other assistant and check the answer carefully: every percentage should be calculated from your numbers, there should be no reasons invented for the changes, and the report should end with questions about what you still need to find out. If a number looks wrong, ask the assistant to show the calculation.

Try it

Monthly report from a table

Frequently asked questions

What is Claude Models on Google Cloud?

It is the Claude API run by Anthropic, reached through a Google Cloud project. Access is managed with Google Cloud IAM and billing goes through Google Cloud Marketplace, according to Anthropic's release notes of 22 October 2026.

Do prompts work differently on Google Cloud?

No. The model is the same Claude, so a prompt that works well on claude.ai or the Claude API should work the same way when your company reaches Claude through Google Cloud.

Is fast mode available?

Yes. Anthropic says fast mode for Claude Opus 5.5 is available on Claude Models on Google Cloud (Anthropic managed).

Sources